Understanding Studies of Resistant Organisms: Focus on Epidemiologic Methods
Robert C. Owens, Lautenbach Ebbing in Antimicrobial Resistance, 2007
Confounding occurs when the association observed between an exposure and outcome is due, in part, to the effect of some other variable. To be a confounder, a variable must be associated with both the exposure and outcome of interest, but cannot be a result of the exposure. Confounding can result in an over- or underestimate of the effect of the exposure of interest. For example, in assessing the association between a FQREC infection and mortality, one must consider underlying severity of illness as a potential confounder. Patients with greater severity of illness are more likely to develop FQREC infection. In addition, greater severity of illness is also more likely to result in mortality. Thus, severity of illness, because it is associated with both the exposure and outcome of interest, is a potential confounding variable. Unlike bias, a confounding variable may be controlled for in the study analysis. However, in order to do this, data regarding the presence or absence of the confounder must be collected during the study.
The Case Control Study: Odds, Odds Ratio – The Concept of Confounding
Johan Giesecke in Modern Infectious Disease Epidemiology, 2017
The proper definition is that a confounder is a factor which is associated with the exposure one is studying and at the same time associated with the outcome. In the fish example, the mayonnaise is the confounder, since it is strongly correlated with the exposure ‘eating fish’ and at the same time with the outcome gastroenteritis. Strictly speaking, the confounder thus need not be the ‘real cause’ of the disease; it could be any exposure that is associated at the same time with another exposure and with the outcome. In the rock concert example above, age is a confounder even though it is not the cause of chlamydia infection. In everyday epidemiological jargon, the term ‘confounder’ is loosely used for those factors that might influence the strength of association (i.e. the magnitude of the RR or the OR) between the risk factor we are presently studying and the disease.
Etiological explanations
Olaf Dammann in Etiological Explanations, 2020
The idea is that randomization renders the exposure independent of all possible endogenous and exogenous confounding influences. Confounders are characteristics of study participants that are associated with both the exposure and the outcome of interest, for example, a common cause. Any confounder can bias the dataset toward or away from “the null,” the numerical measure of “no effect.” Randomization is thought of as ascertaining that all confounders are distributed equally among those who receive the intervention and those who do not. Philosopher John Worrall has argued that randomization cannot guarantee the absence of residual confounding (Worrall 2007) because “the unverifiable assumption of no unmeasured confounding of the exposure effect is necessary for causal inference from observational data” (Hernán and Robins 2006b). And despite all randomization, even RCTs yield only measures of association based on observations. Because “there is no logic that gets you from probabilities or associations alone to causal conclusions” (Kincaid 2011) in observational studies or in RCTs, the latter provide no epistemological impact beyond that provided by the former. Consequently, to consider causal inference based solely on the notion of randomization better than causal inference based on “observational” studies seems to be unjustified.
The risk of cardiovascular disease in women with a history of miscarriage and/or stillbirth
Published in Health Care for Women International, 2019
Farnoosh Asgharvahedi, Leila Gholizadeh, Soraya Siabani
In another study, researchers (Parker et al., 2014) applied a retrospective approach to examine the relationship between miscarriage and later CVD events in 77,701 post-menopausal women, with a mean follow up of 7.7 years. Out of the participants, 30.3% had a history of miscarriage, 2.2% stillbirth, and 2.2% both. The researchers found that women with a history of one miscarriage had greater risk for CHD compared with women without this history, with the multivariable-adjusted odds ratio (OR) of 1.19 (95% CI 1.08–1.32), however, the risk was not significantly different from women who had two or more miscarriages with the reported OR of 1.18 (95% CI 1.04–1.34). There was also no association between miscarriage and future ischemic stroke. In this study, the researchers accounted for the effects of most potential confounders. The researchers hypothesized that metabolic, hormonal, and hemostatic changes associated with pregnancy loss were likely to contribute to the increased risk of CHD in the affected women (Parker et al., 2014).
Risk and cost of infection-related hospitalizations in medicare beneficiaries with comorbid rheumatoid arthritis treated with abatacept versus other targeted disease-modifying anti-rheumatic drugs
Published in Journal of Medical Economics, 2021
Vardhaman Patel, Zulkarnain Pulungan, Anne Shah, Mahesh Kambhampati, Francis Lobo, Allison Petrilla
Some limitations associated with this study and observational studies in general need to be acknowledged. Claims data exist mainly for billing and reimbursement purposes thus, there is a possibility for errors in documentation of medical conditions and outcomes which may lead to patient misclassification either due to miscoding or misdiagnosis. As claims data only captures those disease entities and variables that have their own specific billing codes, this study cannot be used to determine cause and effect. In comparison to cross-sectional and case–control study design, this retrospective cohort study will have higher internal validity. As for all observational studies, treatments are prescribed on the basis of clinical judgment. Therefore, comparisons of patients on different treatment regimens will likely be confounded because there are differences in disease severity and risk of the events of interest (e.g. patients receiving one tDMARD abatacept are likely to be different in many ways from patients receiving other tDMARDs). Potential confounding variables were controlled for via appropriate study design and statistical modelling. However, the possibility of residual confounding from unmeasured factors cannot be excluded. Conclusions regarding the differential risk in serious infections may not be generalizable to specific mechanism of action (MOAs) within the non-TNFi arm. Rituximab and tocilizumab accounted for majority of the treatment within the non-TNFi arm; tofacitinib was used in less than 13% of patients.
The effect of confounding variables in studies of lead exposure and IQ
Published in Critical Reviews in Toxicology, 2020
Cynthia Van Landingham, William G. Fuller, Rosalind A. Schoof
In regression analyses of epidemiological data, there can be confusion regarding which variables are covariates and which are confounders (Gurka 2018). For this analysis we are using the following definitions: a covariate is a variable that correlates with the outcome independent of the major exposure variable, whereas a confounder is a variable that correlates both with the outcome and with the major exposure variable. (Figure 1). Multivariate regression can include both covariates and confounders as independent variables to “correct” for the effect of these variables. In the case of confounders, including the confounder as an independent variable only accounts for the differences in variance and will not account for its effect on the outcome. An interaction term between the confounder and the outcome must be included in the regression model to account for this effect, which is similar to the way a modifier is considered in regression modeling (Gurka 2018). The difference would be that for a confounder, both the independent confounder and the interaction between the confounder and the primary exposure parameter should be included in the model. In the case of the modifier only, the interaction term is necessary.
Related Knowledge Centers
- Causal Inference
- Dependent & Independent Variables
- Correlation Does Not Imply Causation
- Internal Validity
- Conditional Probability
- Randomization
- Bayesian Network
- Collider
- Berkson'S Paradox
- Blocking